Research On Unsupervised Indoor Localization Based On Enhanced Transition Model

With advances in wireless sensors,mobile computing and other technologies,people are increasingly focusing on Location-based services(LBS)and applications.However,the Global Position System(GPS)cannot detect the satellite signal in the area with complex buildings and indoor environment,so the research of indoor positioning technology has been paid great attention.At present,Wi Fi signal is common in indoor environments such as homes and public places.The widespread use of mobile phones and other wearable devices with inertial measurement units(IMU)and Wi Fi modules makes it easy to collect Received Signal Strength(RSS)and IMU data.Therefore,indoor localization methods based on Wi Fi signal and inertial measurement data have attracted many professionals’ research.Compared with the indoor localization method using a single sensor,the localization method using multi-sensor fusion can better meet the current indoor localization needs.In order to solve the problem of high cost of localization method based on Wi Fi signal and the problem of accumulated drift errors of localization method based on Pedestrian Dead Reckoning(PDR),an unsupervised indoor localization method integrating Wi Fi signal and inertial measurement data was proposed.It can achieve the localization effect matching with Wi Fi fingerprint location without too much prior knowledge,but its localization method based on transition model still has many challenges,such as highcost time,poor localization accuracy,etc.In this paper,the traditional unsupervised indoor localization method based on transition model is improved and studied.The main research contents are as follows(1)This paper proposes a novel transition model called Enhanced Transition Model to predict Motion from signal change(ETMM).The ETMM model localization method is divided into two stages.The offline stage collects RSS+IMU crowdsourced trajectory data,and establishes the corresponding relationship between RSS signal changes and motion changes;In the online phase,real-time and offline motion changes are used to realize real-time online localization combined with Kalman filtering.In order to improve the robustness of the design method in this paper,trajectory data enhancement is proposed to enrich the diversity of offline crowdsourced trajectory data,so that the model can learn more comprehensive and detailed information from the environment.In this paper,effective RSS preprocessing is used to reduce the data dimension and data search set,and significantly reduce the computational burden.In order to improve the robustness and accuracy of the model,Direction Matching(DM)constraint was proposed to enhance the mapping relationship between the changes of continuous RSS signals and the changes of motion.Experimental results show that ETMM has better performance than the existing methods in localization accuracy,calculation cost and robustness.In order to improve the robustness and accuracy of the model,Direction Matching(DM)constraint was proposed to enhance the mapping relationship between the changes of continuous RSS signal and the changes of motion.Experimental results show that ETMM has better performance than the existing methods in localization accuracy,calculation cost and robustness.(2)In the online stage of ETMM,Kalman filter algorithm is used to track and locate pedestrians.Kalman filter algorithm has been proved to be able to obtain optimal estimation in linear and Gaussian systems.Although the state equation and observation equation designed in this paper are linear,the mobile phone sensor is vulnerable to the influence of the environment in the process of movement,resulting in the system noise not meeting the Gaussian white noise,resulting in the risk of filtering divergence in ETMM.Particle filter is based on Monte Carlo method and uses particle set to represent the probability.It can approximate the real model in any form of state,and it is widely used in nonlinear non-Gaussian systems.So this paper proposes an enhanced transition model PF-ETMM based on particle filter algorithm.In this paper,two different real environment data sets are used for experimental evaluation,and the effectiveness of particle filter is proved.The localization accuracy of PF-ETMM is 9% higher than that of ETMM.